CN106529530A - Monocular vision-based ahead vehicle detection method - Google Patents
Monocular vision-based ahead vehicle detection method Download PDFInfo
- Publication number
- CN106529530A CN106529530A CN201610965739.5A CN201610965739A CN106529530A CN 106529530 A CN106529530 A CN 106529530A CN 201610965739 A CN201610965739 A CN 201610965739A CN 106529530 A CN106529530 A CN 106529530A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- image
- interest
- haar
- detection method
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The present invention provides a monocular vision-based ahead vehicle detection method and belongs to the vehicle driving assisting field. According to the method, vehicle morphological features and HAAR features are combined, so that the defect of high false negative rate of a method according to which only vehicle morphological features are adopted and the defect of long processing time of a method according to which only HAAR features are adopted can be eliminated; possible regions of interest of vehicles are extracted based on a characteristic that the vehicles on roads have shadows under a lighting condition; the HAAR features of the regions of interest are extracted; a vehicle classifier which is trained based on the Adaboost is adopted to screen the regions of interest; and targets are verified according to a symmetric energy function and an information entropy function. According to the method of the invention, vehicles in road images are marked and highlighted, so that a driver can be prompted to pay attention to distances between his or her own vehicle and vehicles ahead.
Description
Technical field
The present invention relates to machine learning, image procossing, pattern recognition, belong to machine vision and vehicle electric field, especially
It is related to a kind of front vehicles detection method based on monocular vision.
Background technology
In recent years, China's economy is developed rapidly, and automobile occupancy volume is also consequently increased rapidly per capita, but vehicle accident
Quantity be also the trend for increasing and presenting rising year by year.Counted according to 2015, the direct economic loss that vehicle accident is caused
Already exceed 2,000,000,000 yuan.Rear-end collision is occurrence frequency highest in all vehicle accidents, and one of its reason is on highway
Speed is too high；Which two is that driver has found that front collides the higher vehicle of hazard index in time.Intelligent transportation system
(Intelligent Transportation System, ITS) can prevent vehicle accident so as to reduce unnecessary loss,
Therefore ITS has become the focus of global scientific research.If number is it has been found that occur first 0.5 second to driving in vehicle accident according to statistics
Member reports to the police, then can avoid 70% unnecessary vehicle accident.
Front vehicles detecting system is a kind of active safety technologies of vehicle, i.e., before accident is caused danger, to accident
Carry out prevention and accident in real time to evade, and then the generation for avoiding traffic accident.The system be in order to reduce the burden of driver and
A kind of automobile assistant driving system of error in judgement, as driver as carelessness may not find front vehicles, system in time
Prompting driver makes corresponding brake.The real-time road condition information of road ahead is gathered by monocular CMOS camera, is extracted
More specific location information of the front vehicles in collection image, and feed back the safety for improving vehicle traveling in driver.Israel
One image perception technology company Mobileye is directed generally to computer vision algorithms make and the drive assist system of auto industry
Chip technology research, in terms of its product is applied to tesla's pilotless automobile, as the country is to unmanned vapour
The research of car progressively launches, and the research of safety driving assist system there has also been new development.At present, vehicle testing techniques mainly have
Three kinds of sensors is available：Detect based on infrared detection, based on millimetre-wave radar and based on image vision process.The present invention is adopted
It is that, based on image vision treatment technology, the method has the advantages that cost is relatively low, operation strategies are wide and it is convenient to implement,
Therefore it is with a wide range of applications.The domestic paper and patent documentation for front vehicles detection technique also being had to many correlations：
Saic Chery Automobile Co., Ltd's invention《Front vehicles detection method》(application number：201510340493.8), propose a kind of
Combined according to vision-based detection and radar detection and obtain vehicle position information, the invention accuracy is higher and can solve reason
Weeds interference nearby, improves the accuracy of vehicle detection, but make use of two sensors of radar detection and vision-based detection, equipment
It is relatively costly；The invention that Chang An University releases《A kind of detection method of highway front vehicles》(application number：
201510296145.5) label is carried out by the testing result image to vehicle's contour, obtains label vehicle；Scan grades figure
Picture, calculates the extraneous rectangle of each label vehicle according to extraneous rectangle is obtained, and calculates the extraneous rectangle ginseng of each label vehicle
Number, it is easier to detect vehicle according to this vehicle morphological characteristic of rectangle parameter of the vehicle external world, but loss and false drop rate
Comparison is high.
Consult the technical brief discovery of related fields, two primary evaluations of front vehicles detection method based on monocular vision
Real-time and the accuracy of vehicle identification of the index for the image procossing of system, this performance indications are influenced each other, are mutually restricted.
So, a kind of preferable vehicle detecting system should be able to balance real-time and accuracy.
The content of the invention
For the defect that current techniques are present, it is an object of the invention to provide a kind of front vehicles based on monocular vision
Detection method, its accuracy in detection are high, and system real time can be good, can drop as far as possible under the precondition of accurate vehicle
The complexity of low algorithm, reduces the recognition time of system, is allowed to be extended to vehicle registration carrier and vehicle adaptive navigation.Specifically
For be that one kind is detected to front vehicles according to vehicle bottom shadow character and with reference to HAAR features, according to what is detected
Vehicle provides corresponding information to driver.
It is for reaching above-mentioned purpose, as follows to vehicle detecting algorithm basic ideas in the present invention：
Realize to vehicle detection with reference to vehicle morphological characteristic (bottom has shade) and HAAR features.It is special by vehicle form
Acquisition area-of-interest is levied, then in the HAAR features for extracting area-of-interest, compared to the HAAR of extracting directly entire image
Feature, can substantially reduce the dimension of feature, reduce the time of target recognition, combined using morphological characteristic and HAAR features
Thinking can also improve the accuracy of algorithm.
According to above-mentioned design, the present invention is adopted the following technical scheme that：
Present example provides a kind of vehicle checking method based on monocular vision, and the step includes：
(1) image acquisition：CMOS camera is installed to obtain road ahead scene above Herba Plantaginis, photographic head installs the time
The position that axle is as parallel to the ground as possible and photographic head is installed must be changeless；Next adjusts photographic head and Jiao
Away from there is error in reducing follow-up image procossing.
(2) Image semantic classification：First the original image that imageing sensor is obtained is cut out, above image it is most of all
It is sky, the possibility that there is no vehicle appearance then only retains 2/3 part of lower section of original image, reduces image procossing
Data volume；Then to cutting out after image carry out gray processing, using the method for weighted intensity；Afterwards in order to effective filter out arteries and veins
Rush noise and make object edge still than more visible, gray level image is filtered using median filter method；Finally adopt
Adaptive approach carries out binaryzation to filtered image twice, by counting average gray value and the standard of road surface pixel
Difference, the optimal threshold for obtaining can the vehicle shadow on road surface from background from separating, this kind of method can be taken into account and be
The robustness and real-time of system.
(3) morphological image process：The behaviour such as the isolated cavity of burn into expansion and filling are carried out to pretreated image
Make, so preliminarily target can be separated from each other with background.
(4) area-of-interest (Regions of Interest, ROI) is extracted：The preliminary area-of-interest for extracting of the invention
It is that shade this feature is had under normal lighting conditions according to vehicle bottom, so as to realize to being some of vehicle region
Separate from artwork part.In order to reduce interference, first from top to bottom, search for from left to right hatched initial position and
Final position, obtains according to camera position and parameter calibration, and the hatched scope of underbody can change with the difference being expert at,
Then system sets a threshold value to every a line, assert that this line is underbody if the shade line length of detection is close with threshold value
Hacures, per the corresponding length threshold relation of a line be：
Wherein w is hacures length in the picture；wpFor vehicle developed width；H be CMOS camera optical axis from the ground
Actual height；Y be required hacures longitudinal axis line number in the picture；H is picturedeep.
(5) HAAR features are extracted：Area-of-interest to separating is normalized, i.e., size is identical, then
Extract corresponding five classes HAAR feature and preserve.
(6) target recognition：The HAAR features for extracting, cascade out a strong classifier pair using Adaboost algorithm
The carrying out of area-of-interest is classified, and screens above all area-of-interests, then records corresponding coordinate points if vehicle region
And width.
Above-mentioned target recognition realized according to HAAR+Adaboost algorithms, area-of-interest is carried out point based on this
Class is screened, specifically comprising following step：
Step one：Positive negative sample is opened, vehicle in positive sample, is only included, negative sample is environment beside road, then sample
This size normalization is 32*32；
Step 2：Five kinds of HAAR features of positive negative sample are extracted, five kinds of HAAR features are together in series composition sample image
HAAR characteristic vectors；
Step 3：The HAAR characteristic vectors obtained in step 2 are imported in Matlab, using Adaboost graders
The HAAR features for aligning negative sample carry out off-line training；
Step 4：After the positive and negative sample training of certain amount, a strong classifier is cascaded into, then area-of-interest
It is normalized size and extracts corresponding HAAR features, area-of-interest is classified based on HAAR features.
(7) target verification：Construction odd even energy function and information flow function, the vehicle region to identifying are carried out symmetrically
Property checking and comentropy checking, improve vehicle detecting algorithm accuracy rate.Screened using vehicle morphological feature, due to car
Afterbody has texture and certain symmetry, and use information entropy function test-target region is with the presence or absence of abundant texture letter
Breath；Edge treated upright projection is carried out to target using Sobel operators, then constructs energy function inspection symmetry.
For realizing said method, the invention provides a kind of front vehicles detecting system, key component includes：Image
Sensor, processor, display and data storage.Wherein imageing sensor, display and processor connection.Imageing sensor
It is, for being acquired to front reason image, to obtain original image；Processor is that the original image for collecting is carried out image
Process, there will be vehicle region and be partially separated out；Display is for the car for showing original image in real time and outline in artwork
, to remind human pilot；Memorizer is that compiling Intel officials carry in order to (SuSE) Linux OS is built in face on a processor
For OpenCV visions storehouse.
Preferably, the imageing sensor that uses of the present invention is vehicle-mounted CMOS camera, road directed straight ahead, installation site
For the windshield of vehicle front, it is about 1 meter or so apart from ground level；The processor that the present invention is used is Xilinx's
Zynq-zc702 development boards；The memorizer that the present invention is used is SD card, inserts SDIO；The display that the present invention is used is high definition car
HDMI display is carried, is connected with exploitation plate interface by HDMI wire.
Compared with existing vehicle testing techniques, the present invention has following prominent substantive distinguishing features excellent with significant
Point：
A) it is easy for installation, low cost.The sensor that present invention design needs is less, it is only necessary to which single CMOS camera just can
The collection to road environment is realized, using Vehicular display device, integral layout is relatively simplified, be easy to realize independently all kinds of cars
Install.
B) reliability is high.Using vehicle morphological characteristic and HAAR features can well by vehicle detection out, merge this
Two category features, improve capacity of resisting disturbance and the real-time performance of vehicle detecting system.
C) debug convenient.Can bring many convenient with OpenCV on Zynq platforms, realize on Zynq platforms from
Photographic head real-time monitoring and the position for marking vehicle, while realize shown to the picture before labelling and after labelling in real time
And deposit, developed on OpenCV programs and PC platforms on the platform and there is no difference, can be realized on PC to calculating
The emulation of method is simultaneously debugged, and just directly can be run on Zynq platforms after cross compile.
Description of the drawings
Fig. 1 is the inventive method embodiment flow chart.
Fig. 2 is that embodiment of the present invention SD card subregion builds linux system schematic diagram.
Fig. 3 is embodiment of the present invention chip ADV7511 functional structure charts.
Fig. 4 is embodiment of the present invention system hardware Organization Chart.
Fig. 5 is embodiment of the present invention SD card interface circuit schematic diagram.
Specific embodiment
In order to be well understood to the technology of the present invention embodiment, below in conjunction with accompanying drawing, the present invention is further elaborated says
It is bright.
As shown in figure 1, a kind of front vehicles detection method based on monocular vision, including following several steps：
(a) image acquisition：Vehicle front road scene is obtained by fixedly mounting CMOS camera on vehicle；
(b) pretreatment：Image to gathering carries out cutting, gray processing, removes noise and binarization operation；
(c) Morphological scale-space：Burn into expansion and the isolated cavity operation of filling is carried out to pretreated image；
D () area-of-interest, i.e. Regions of Interest, ROI are extracted：Shadow character is had according to vehicle bottom
This characteristic, is that the part of vehicle region separates to probability；
E () extracts HAAR features：Formed objects are normalized to ROI, then extract the corresponding HAAR features of ROI simultaneously
Preserve；
(f) target recognition：A strong classifier is cascaded by HAAR features with reference to Adaboost to classify ROI, it is right
Vehicle is identified；
(g) target verification：ROI to identifying carries out symmetry checking and comentropy checking, improves vehicle detection and calculates
The accuracy rate of method.
In this example, the image for gathering from monocular cam is through cutting out, after gray processing and filtering, using self adaptation
Algorithm carries out binaryzation, as the hacures of vehicle are easily affected by background and illumination, by using adaptive thresholding twice
Value segmented extraction underbody shade, can eliminate external interference, and robustness is preferable.In first time Threshold segmentation rejection image brightness compared with
The impact of high pixel, can exclude the factors such as light intensity and cause the higher situation of picture quality brightness value, the second subthreshold point
Cutting the scope asked for is not entire image but the pixel of threshold value that splits less than first time of gray value, to less than threshold value
Pixel calculate threshold value using adaptive polo placement formula.
Wherein, gray values of the f (m, n) for image；The height and width of M and N difference representative images；μ is pixel gray level
The meansigma methodss of value；σ represents the variance of pixel gray value；Threshold is the threshold value of black and white segmentation.Extract cloudy using the algorithm
Hachure can preferably retain underbody shadow information, effectively suppress ambient interferences and illumination condition to extracting hatched shadow
Ring.
In this example, to it is hatched extract be by entire image scanning (from top to bottom, from left to right) search for
Beginning position and final position, it is then determined that hacures coordinate in the picture and length.Can be with according to image image-forming principle
It was found that, the distance between vehicle-mounted camera and front vehicles are reflected on image the ratio taken in entire image for vehicle.When
When spacing is more remote, underbody hacures are shorter；When vehicle is nearer, underbody hacures are longer.I.e. in the ordinate of orthogonal axes of image, can
To set the threshold value of a shade line length to every a line, when the shade line length for detecting exceedes or be less than the row threshold value,
The partial phantom line is then deleted, hacures are then merged, it is ensured that hatched left end point is close to target left-hand point after merging,
Hatched right endpoint is close to target right-hand point.According to the principle vehicle of pinhole imaging system length w in the picture and vehicle reality
Width ws, the photographic head optical axis distance ground level H and hacures coordinate y on image y direction have following relation：
In this example, need to carry out these area-of-interests intelligent classification, the calculation of employing after extracting area-of-interest
Method is HAAR+Adaboost.Positive sample and negative sample is made first, and positive and negative samples pictures size normalization is 32 × 32, wherein
The picture of vehicle is only included in positive sample, can not be there is vehicle and be ensured the multiformity of negative sample, as far as possible not in negative sample
To repeat, using integrogram, extract five kinds of HAAR features of positive negative sample respectively, then the characteristic vector of all pictures is connected
To constitute the HAAR characteristic vectors of positive negative sample；Then characteristic vector is imported in MATLAB, using Adaboost algorithm to spy
Levying vector carries out off-line training, obtains the vehicle classification device for target recognition；Next area-of-interest is normalized
It is in the same size with sample image, the HAAR features of area-of-interest are extracted, is finally realized to region of interest using vehicle classification device
The classification in domain.
Substantially vehicle can be identified from background by above step, but may wherein there are some mistakes
The part of inspection, the present invention are further verified to vehicle region.Vehicle according to vehicle have obvious symmetry this effectively according to right
Classification target out is verified.Vertical edge enhancing figure is obtained to target process initially with Sobel vertical edge operators
Picture, then calculates the symmetry of vertical edge, by the vertical direction Vertical edge projection average in square frame as abscissa
One-dimensional functions g (x), according to mathematical theory, any one one-dimensional functions can be synthesized by an even function and an odd function, with
Mid-point is the longitudinal axis, and the odd function component and even function component of one-dimensional functions are respectively：
Wherein, even function E>0, then its average is certainly more than 0, and the average of odd function O is 0, it is ensured that even function it is equal
It is worth for 0, dual function component E is normalized：
The energy of the even function component by odd function component and after normalized represents profile symmetry, strange letter
Several energy is bigger, shows that symmetry is less, and the energy of even function is bigger, shows that symmetry is higher.Target vertical direction is symmetrical
Property is estimated：
S ∈ [- 1,1] in formula, it is as S=-1, completely asymmetric；It is as S=1, full symmetric.It is found through experiments and works as S>
Can think substantially symmetrical when 0.3.Test result shows that the interval can effectively exclude false vehicle.
Comentropy h (x) is defined as the mathematic expectaion of quantity of information.H (x) gray level images span is [0, Hmax].Usual feelings
Under condition, road surface intensity profile is more uniform, and entropy is less, but the quantity of information that vehicle contains is relatively more, vehicle rearview window, car plate,
The informative that bumper etc. is included, entropy are relatively large.
P (l in formulai) it is frequency that certain pixel occurs, rows represents the line number of region of interest area image.A large amount of vehicles are entered
Row statistics, finds to work asThis interval interior energy effectively excludes false vehicle.
For realizing said method, the present invention is used and is closely integrated with FPGA with ARM Cortex A9 stones
To Zynq-zc702 development boards together.By such combination, processor was both played and had processed complex control algorithm, operation
The advantage of the aspects such as operating system, make use of FPGA parallel algorithm accelerate, can dynamic recognition the characteristics of, improve system
Motility while realize hardware-accelerated, also accelerate the development rate of hardware.The process of hardware algorithm is accelerated, while
The complete independently Hardware/Software Collaborative Design on Zynq.The architecture platform of FPGA+ARM has high speed, the double dominant of easy exploiting, is
Realize the excellent selection of high definition, scan picture.
First have to transplant (SuSE) Linux OS in zc702 and compile OpenCV image procossings storehouse before algorithm is realized.It is accurate
A standby capacity is the SD card of 8GB, is divided into Liang Ge areas to SD card, as shown in Fig. 2 first subregion is that FAT forms are used for placing
Bootloader, device tree and kernel mirror image, second subregion is EXT4 forms, for placing linaro-Linux files system
System.Next compiling configuration u-boot, configures binary file system.bit used by FPGA, generates System startup files
BOOT.BIN, compiles linux kernel, and compiling is supported HDMI outut device tree source files and supports the device tree of FMC video frequency outputs
Source file.After putting up linaro-Linux platforms, cross compile and installation are completed to OpenCV source codes, be so easy to build base
In the Video processing software platform of monocular vision.
Mainly HDMI (HDMI) is designed in this example.The framework of HDMI systems is by believing
Place and information source are constituted.In HDMI message transmitting procedures, three phases can be divided into：Video data transmission cycle, controlling transmission
Cycle and data islands transmission cycle.What is transmitted in video frequency output transmission cycle is valid pixel information, in data islands transmission week
In phase, transmission is the audio frequency and additional data packed, and is controlling cycle in data transfer task above.Zc702's
HDMI uses a ADV7511 of ADI companies production, the HDMI transmitters of 225MHz.As shown in figure 3, main wrap
Containing following functional device：HDMI Ethernet passages, audio data receipt module, video data are received and compression module,
High broadband digital content protection module and TMDS output modules.4 pairs of differential signals of ADV7511 input to the output section of HDMI
Point, then exported.The system design figure of Zynq as shown in figure 4, need to select suitable IP kernel using ADV7511, so
Complete SoC systems could be constituted.
In the present embodiment, CMOS camera interface uses the USB OTG DLLs on zc-702.Development board makes
With usb0 control ports as USB OTG interfaces, outside employs the physical layer transmission chip TUSB1210 of TI companies, is used for
The MIO pins of PS parts in connection Zynq.When USB device can not receive the port USB power source voltage, as usb host mouth
With can be provided out 5v voltages during OTG.
In the present embodiment, built-in digital input/output (SDIO) interface of zc702 development boards, the interface is use
Family provides and accesses not volatibility SD storage card ancillary equipment.As shown in figure 5, SDIO port signals are by being connected to SoC systems
PS parts, and cause the level of VCCMIO to be pulled to 1.8V, expanded using SDIO ports between XC7Z020AP SoC and SD card seat
Exhibition device, the expander have used voltage conversion technology.Linux development platform is built well can using SD card memorizer.
Claims (6)
1. a kind of front vehicles detection method based on monocular vision, it is characterised in that including following several steps：
(a) image acquisition：Vehicle front road scene is obtained by fixedly mounting CMOS camera on vehicle；
(b) pretreatment：Image to gathering carries out cutting, gray processing, removes noise and binarization operation；
(c) Morphological scale-space：Burn into expansion and the isolated cavity operation of filling is carried out to pretreated image；
D () area-of-interest, i.e. Regions of Interest, ROI are extracted：According to vehicle bottom exist shadow character this
Characteristic, the part to being probably vehicle region are separated；
E () extracts HAAR features：Formed objects are normalized to ROI, are then extracted the corresponding HAAR features of ROI and are protected
Deposit；
(f) target recognition：A strong classifier is cascaded by HAAR features with reference to Adaboost to classify ROI, to vehicle
It is identified；
(g) target verification：ROI to identifying carries out symmetry checking and comentropy checking, improves vehicle detecting algorithm
Accuracy rate.
2. the front vehicles detection method based on monocular vision according to claim 1, it is characterised in that the step
A the position for installing photographic head in () is fixed, and adjust fixed height and fixed focal length, to reduce at successive image
Occurs error in reason.
3. the front vehicles detection method based on monocular vision according to claim 1, it is characterised in that the step
B, in () pretreatment, it is to retain 2/3 part below collection image to cut out, to reduce data processing amount；Gray processing adopts weighted intensity
Change；Method of the noise using medium filtering is removed, to effective filter out impulsive noise and make object edge still than more visible；
Binary Sketch of Grey Scale Image is using adaptive threshold fuzziness twice, by counting average gray value and the standard of road surface pixel
Difference, obtains optimal threshold, can the vehicle shadow on road surface from background from separating, and the robust of system can be taken into account
Property and real-time.
4. the front vehicles detection method based on monocular vision according to claim 1, it is characterised in that the step
D the area-of-interest extracted in () is the image-region that there is shade, in order to reduce interference, first from top to bottom, past from a left side
It is right to search for hatched initial position and final position, obtained according to camera position and parameter calibration, the hatched model of underbody
Enclose and can change with the difference being expert at, then system sets a threshold value to every a line, if the shade line length of detection and threshold
Value is close then to assert that this line is underbody hacures, per the corresponding length threshold relation of a line is：
Wherein w is hacures length in the picture；wpFor vehicle developed width；H is CMOS camera optical axis reality from the ground
Highly；Y be required hacures longitudinal axis line number in the picture；H is picturedeep.
5. the front vehicles detection method based on monocular vision according to claim 1, it is characterised in that the step
E () and (f) target recognition realized according to HAAR+Adaboost algorithms, carry out classifying screen based on this to area-of-interest
Choosing, wherein comprising following step：
Step one：Positive negative sample is opened, vehicle in positive sample, is only included, negative sample is environment beside road, then sample
Size normalization is 32*32；
Step 2：Five kinds of HAAR features of positive negative sample are extracted, five kinds of HAAR features is together in series and is constituted sample image
HAAR characteristic vectors；
Step 3：The HAAR characteristic vectors obtained in step 2 are imported in Matlab, is aligned using Adaboost graders
The HAAR features of negative sample carry out off-line training；
Step 4：After the positive and negative sample training of certain amount, a strong classifier is cascaded into, then area-of-interest is carried out
Normalization size simultaneously extracts corresponding HAAR features, area-of-interest is classified based on HAAR features.
6. the front vehicles detection method based on monocular vision according to claim 1, it is characterised in that the step
G () is screened using vehicle morphological feature, as vehicle tail has texture and certain symmetry, use information entropy letter
Number test-target region is with the presence or absence of abundant texture information；Edge treated is carried out to target using Sobel operators and is vertically thrown
Shadow, then constructs energy function inspection symmetry, improves the accuracy rate of front vehicles detecting system.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610965739.5A CN106529530A (en) | 2016-10-28 | 2016-10-28 | Monocular vision-based ahead vehicle detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610965739.5A CN106529530A (en) | 2016-10-28 | 2016-10-28 | Monocular vision-based ahead vehicle detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106529530A true CN106529530A (en) | 2017-03-22 |
Family
ID=58327055
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610965739.5A Pending CN106529530A (en) | 2016-10-28 | 2016-10-28 | Monocular vision-based ahead vehicle detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106529530A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951879A (en) * | 2017-03-29 | 2017-07-14 | 重庆大学 | Multiple features fusion vehicle checking method based on camera and millimetre-wave radar |
| CN107122734A (en) * | 2017-04-25 | 2017-09-01 | 武汉理工大学 | A kind of moving vehicle detection algorithm based on machine vision and machine learning |
| CN107643295A (en) * | 2017-08-24 | 2018-01-30 | 中国地质大学（武汉） | A kind of method and system of the cloth defect on-line checking based on machine vision |
| CN107875638A (en) * | 2017-11-24 | 2018-04-06 | 宁波隆翔环保科技有限公司 | One kind game car |
| CN108133231A (en) * | 2017-12-14 | 2018-06-08 | 江苏大学 | A kind of real-time vehicle detection method of dimension self-adaption |
| CN108460323A (en) * | 2017-12-29 | 2018-08-28 | 惠州市德赛西威汽车电子股份有限公司 | A kind of backsight blind area vehicle checking method of fusion vehicle mounted guidance information |
| CN109146807A (en) * | 2018-07-31 | 2019-01-04 | 南昌工程学院 | The rapid detection method of vehicle in a kind of traffic video |
| CN109910790A (en) * | 2019-03-05 | 2019-06-21 | 同济大学 | A kind of ADAS domain controller |
| CN110502971A (en) * | 2019-07-05 | 2019-11-26 | 江苏大学 | Road vehicle recognition methods and system based on monocular vision |
| CN110765929A (en) * | 2019-10-21 | 2020-02-07 | 东软睿驰汽车技术（沈阳）有限公司 | Vehicle obstacle detection method and device |
| CN111126224A (en) * | 2019-12-17 | 2020-05-08 | 成都通甲优博科技有限责任公司 | Vehicle detection method and classification recognition model training method |
| CN112255641A (en) * | 2020-09-30 | 2021-01-22 | 成都新成汽车检测设备有限公司 | Method for measuring automobile wheel base |
| CN113723282A (en) * | 2021-08-30 | 2021-11-30 | 上海商汤临港智能科技有限公司 | Vehicle driving prompting method and device, electronic equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7366323B1 (en) * | 2004-02-19 | 2008-04-29 | Research Foundation Of State University Of New York | Hierarchical static shadow detection method |
| CN103150560A (en) * | 2013-03-15 | 2013-06-12 | 福州龙吟信息技术有限公司 | Method for realizing intelligent safe driving of automobile |
| US20130182904A1 (en) * | 2012-01-17 | 2013-07-18 | Objectvideo, Inc. | System and method for video content analysis using depth sensing |
| CN104392212A (en) * | 2014-11-14 | 2015-03-04 | 北京工业大学 | Method for detecting road information and identifying forward vehicles based on vision |
| CN104866823A (en) * | 2015-05-11 | 2015-08-26 | 重庆邮电大学 | Vehicle detection and tracking method based on monocular vision |
| CN105206109A (en) * | 2015-08-13 | 2015-12-30 | 长安大学 | Infrared CCD based foggy day identifying early-warning system and method for vehicle |
-
2016
- 2016-10-28 CN CN201610965739.5A patent/CN106529530A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7366323B1 (en) * | 2004-02-19 | 2008-04-29 | Research Foundation Of State University Of New York | Hierarchical static shadow detection method |
| US20130182904A1 (en) * | 2012-01-17 | 2013-07-18 | Objectvideo, Inc. | System and method for video content analysis using depth sensing |
| CN103150560A (en) * | 2013-03-15 | 2013-06-12 | 福州龙吟信息技术有限公司 | Method for realizing intelligent safe driving of automobile |
| CN104392212A (en) * | 2014-11-14 | 2015-03-04 | 北京工业大学 | Method for detecting road information and identifying forward vehicles based on vision |
| CN104866823A (en) * | 2015-05-11 | 2015-08-26 | 重庆邮电大学 | Vehicle detection and tracking method based on monocular vision |
| CN105206109A (en) * | 2015-08-13 | 2015-12-30 | 长安大学 | Infrared CCD based foggy day identifying early-warning system and method for vehicle |
Non-Patent Citations (7)
| Title |
|---|
| KIM, MS等: "On road vehicle detection by learning hard samples and filtering false alarms from shadow features", 《JOURNAL OF MECHANICAL SCIENCE AND TECHNOLOGY》 * |
| LI, WH等: "Co-training Algorithm Based on On-line Boosting for Vehicle Tracking", 《2013 IEEE INTERNATIONAL CONFERENCE ON INFORMATION AND AUTOMATION (ICIA)》 * |
| 严泰来 等: "《遥感技术与农业应用》", 31 July 2008 * |
| 赵池航 等: "《交通信息感知理论与方法》", 30 September 2014 * |
| 韩飞龙 等: "一种新的车辆辅助驾驶动态障碍物检测与分类方法", 《计算机应用研究》 * |
| 顾兆伦等: "多环境下的实时前车检测与车距测量", 《信号处理》 * |
| 高磊 等: "基于边缘对称性的视频车辆检测算法", 《北京航空航天大学学报》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951879A (en) * | 2017-03-29 | 2017-07-14 | 重庆大学 | Multiple features fusion vehicle checking method based on camera and millimetre-wave radar |
| CN106951879B (en) * | 2017-03-29 | 2020-04-14 | 重庆大学 | Multi-feature fusion vehicle detection method based on camera and millimeter wave radar |
| CN107122734A (en) * | 2017-04-25 | 2017-09-01 | 武汉理工大学 | A kind of moving vehicle detection algorithm based on machine vision and machine learning |
| CN107643295B (en) * | 2017-08-24 | 2019-08-20 | 中国地质大学（武汉） | A kind of method and system of the cloth defect on-line checking based on machine vision |
| CN107643295A (en) * | 2017-08-24 | 2018-01-30 | 中国地质大学（武汉） | A kind of method and system of the cloth defect on-line checking based on machine vision |
| CN107875638A (en) * | 2017-11-24 | 2018-04-06 | 宁波隆翔环保科技有限公司 | One kind game car |
| CN108133231A (en) * | 2017-12-14 | 2018-06-08 | 江苏大学 | A kind of real-time vehicle detection method of dimension self-adaption |
| CN108460323A (en) * | 2017-12-29 | 2018-08-28 | 惠州市德赛西威汽车电子股份有限公司 | A kind of backsight blind area vehicle checking method of fusion vehicle mounted guidance information |
| CN108460323B (en) * | 2017-12-29 | 2022-05-20 | 惠州市德赛西威汽车电子股份有限公司 | Rearview blind area vehicle detection method fusing vehicle-mounted navigation information |
| CN109146807A (en) * | 2018-07-31 | 2019-01-04 | 南昌工程学院 | The rapid detection method of vehicle in a kind of traffic video |
| CN109146807B (en) * | 2018-07-31 | 2021-04-06 | 南昌工程学院 | Method for rapidly detecting vehicles in traffic video |
| CN109910790A (en) * | 2019-03-05 | 2019-06-21 | 同济大学 | A kind of ADAS domain controller |
| CN109910790B (en) * | 2019-03-05 | 2021-11-09 | 同济大学 | ADAS domain controller |
| CN110502971A (en) * | 2019-07-05 | 2019-11-26 | 江苏大学 | Road vehicle recognition methods and system based on monocular vision |
| CN110765929A (en) * | 2019-10-21 | 2020-02-07 | 东软睿驰汽车技术（沈阳）有限公司 | Vehicle obstacle detection method and device |
| CN111126224A (en) * | 2019-12-17 | 2020-05-08 | 成都通甲优博科技有限责任公司 | Vehicle detection method and classification recognition model training method |
| CN112255641A (en) * | 2020-09-30 | 2021-01-22 | 成都新成汽车检测设备有限公司 | Method for measuring automobile wheel base |
| CN113723282A (en) * | 2021-08-30 | 2021-11-30 | 上海商汤临港智能科技有限公司 | Vehicle driving prompting method and device, electronic equipment and storage medium |
| CN113723282B (en) * | 2021-08-30 | 2024-03-22 | 上海商汤临港智能科技有限公司 | Vehicle driving prompting method, device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106529530A (en) | Monocular vision-based ahead vehicle detection method | |
| CN106169244B (en) | Device and method is provided using the guidance information of crossing recognition result | |
| CN102765365B (en) | Pedestrian detection method based on machine vision and pedestrian anti-collision warning system based on machine vision | |
| CN103034843B (en) | Method for detecting vehicle at night based on monocular vision | |
| CN101587544B (en) | Based on the carried on vehicle antitracking device of computer vision | |
| CN105260712A (en) | Method and system for detecting pedestrian in front of vehicle | |
| CN106127107A (en) | The model recognizing method that multi-channel video information based on license board information and vehicle's contour merges | |
| CN105574552A (en) | Vehicle ranging and collision early warning method based on monocular vision | |
| CN107886034B (en) | Driving reminding method and device and vehicle | |
| CN102637257A (en) | Video-based detection and recognition system and method of vehicles | |
| CN101900567A (en) | No-texture clear path detection based on pixel | |
| CN110210474A (en) | Object detection method and device, equipment and storage medium | |
| CN103984917A (en) | Multi-feature nighttime vehicle detection method based on machine vision | |
| JP7280874B2 (en) | Solid-state image sensor, imaging device, and control method for solid-state image sensor | |
| CN107609472A (en) | A kind of pilotless automobile NI Vision Builder for Automated Inspection based on vehicle-mounted dual camera | |
| CN106803262A (en) | The method that car speed is independently resolved using binocular vision | |
| CN105893960A (en) | Road traffic sign detecting method based on phase symmetry | |
| Li et al. | A low-cost and fast vehicle detection algorithm with a monocular camera for adaptive driving beam systems | |
| Wen et al. | A rear-vehicle detection system for static images based on monocular vision | |
| CN105512662A (en) | Detection method and apparatus for unlicensed vehicle | |
| Chen | Road vehicle recognition algorithm in safety assistant driving based on artificial intelligence | |
| CN112669615B (en) | Parking space detection method and system based on camera | |
| Chang et al. | Stereo-based object detection, classi? cation, and quantitative evaluation with automotive applications | |
| CN114119955A (en) | Method and device for detecting potential dangerous target | |
| Álvarez et al. | Perception advances in outdoor vehicle detection for automatic cruise control |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170322 |